Abstract
Vast amounts of remotely sensed data from a variety of remote sensors deployed on satellites and other platforms are collected regularly. This includes satellite images, aerial images, and Light Detection And Ranging (LiDAR) data. The continuous advancements of the remote sensing systems offered new enhanced data specifications that need evolving processing techniques to exploit the new abilities, resolutions, and accuracies in highly automatic fashion to overcome the slow costly yet accurate human processing.
The main objective of this research is to develop a framework for classification of data acquired from different sensors in object based fashion using a flexible classification engine.
A segmentation approach has been proposed where many similarity measures can be used interchangeably or collaboratively. This proposed approach minimizes the needed parameters and avoids the commonly used assumption of the same scale objects. Multiple data layers from different sources and specifications could be employed collaboratively in this approach with the ability of relative weighting of these layers to represent the relevance of each data layer. The performance of this approach has been compared to human segmentation for assessment. The achieved performance shows the significance of the proposed approach.
Inspired by the flexibility and expandability of human classification abilities, A flexible classification methodology is proposed that can learn through reference data sets, employ rules in accumulative fashion, tolerate missing features, and extend to include new classes. An adaptive sequential classification approach has been proposed to further improve the classification performance.
Several tests of different data sources including satellite images, aerial images, and LiDAR data have been conducted using the proposed classification methodologies. The achieved classification result demonstrates the significance of the proposed segmentation and classification approaches.
